AI NewsGo-to-marketAnnouncement
Bing adds separate publisher controls for AI training and AI grounding
Microsoft split Bing's AI content controls into two separate choices. One is for AI training. One is for AI grounding, which feeds Copilot and AI Overview.

Image: Microsoft Bing
Why it mattersA site can now block Bing training and still appear in Copilot or AI Overview answers. The two choices are set apart.
A site that lets Bing crawl for search but does not want to feed AI training can now say so. Two separate settings now cover the choice. Microsoft split its AI content controls into two choices on 30 October 2026. One governs AI training. The other governs AI grounding. Krishna Madhavan, Principal Product Manager, Microsoft Bing, posted the news on the Bing Blogs.
Grounding is Microsoft's term for the use of a site's content to produce AI answers. Training is the use of content to teach a model. Microsoft said the old NoArchive and NoCache tags are retired. The two new controls replace them.
Three options, set apart
Madhavan listed the choices. AI Training has two values: allow by default, or disallow. AI Grounding has three values. Allow is the default. DisallowExtended keeps a site out of third-party AI answers but allows Microsoft's own products. Disallow blocks all grounding use.
Madhavan wrote what DisallowExtended covers. It limits grounding to Microsoft-operated products. The post names four: AI Mode, AI Overview, Copilot, and Microsoft 365. A site that sets DisallowExtended still appears in those four. Other AI tools that use Bing data do not.
A site can also mix the two settings. Madhavan said a publisher can block training and keep grounding on. The post states this directly. The site stays in AI answers while declining training.
Where to set it
Publishers set the preference in three places. In robots.txt, the training rule uses a new user-agent token: __usage_ai-training. The post gives no sample for the grounding token in robots.txt. In meta tags or HTTP headers, the page-level directives are ai-training:disallow, ai-grounding:disallow, and ai-grounding:disallowextended.
Microsoft said the page-level setting wins. A new AI Training preference set at the page level overrides the robots.txt setting. A site can keep a wide allow in robots.txt and refuse on specific pages.
The dates
Microsoft gave four dates for the roll-out.
- 15 December 2026: temporary grounding preferences open in Bing Webmaster Tools.
- 1 February 2027: Microsoft honours the new preferences during crawling.
- 1 March 2027: deadline for publishers to move off the old directives.
- 30 September 2027: the temporary preferences in Bing Webmaster Tools expire.
Traditional Bing rankings are unaffected by any choice. Blocking grounding reduces visibility in AI answers only.
How this fits with Google's news today
Google also announced plans for a shared AI training control today. Both Microsoft and Google linked their move to the IETF AIPREF working group. Microsoft's post names its own directives rather than the AIPREF draft token.
For a team that owns the site's robots.txt, the useful preparation is one list. Which AI products should the site feed, and which should it decline. The answer sets the Bing values. When the AIPREF line is live, the same answer sets the shared line that other providers will read. The old NoArchive and NoCache tags should come out at the same time, before the 1 March 2027 deadline.
Source
Your Content, Your Choice: Clearer Preferences for Training and Grounding, by Krishna Madhavan, Principal Product Manager, Microsoft Bing. The story was also written up by Barry Schwartz at Search Engine Roundtable.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.


